JP5934031B2 - Gradation reduction apparatus and program - Google Patents

Description

  The present invention relates to a gradation reduction device for reducing gradation of an original image and a program thereof.

  Human visual sensitivity is low for high frequencies where the brightness changes rapidly, and high for low frequencies where it changes slowly. Conventionally, a gradation reduction device that performs gradation reduction (requantization) using a color table detected based on such human visual characteristics is known (for example, see Patent Document 1). More specifically, the gradation reduction device described in Patent Document 1 first uses a reference color table for multi-gradation image data obtained by returning multi-gradation image data to a state in which gamma correction processing has not been performed. To perform gradation reduction processing. The error detection circuit detects the gradation reduction error of the gradation reduced image data and supplies it to the control circuit. The control circuit compares the gradation reduction error with a predetermined threshold value.If the gradation reduction error is equal to or greater than the threshold value, the reference color table is changed until the gradation reduction error is equal to or less than the predetermined threshold value, and the gradation reduction process is performed again. Each unit is controlled to detect a gradation reduction error. Then, the gradation reduction processing of the image is performed using the reference color table when the gradation reduction error is equal to or less than a predetermined threshold.

  In addition, the Lloyd-Max method is known, in which a histogram showing the frequency for each gradation value (luminance value) of an image is examined, and when gradation reduction is performed, more gradations are assigned to gradation values with higher frequencies. (For example, see Non-Patent Documents 1 and 2).

Japanese Patent Laid-Open No. 9-321988

S.P.Lloyd, “Least squares quantization in PCM”, IEEE Trans. Information Theory, vol.IT-28, pp.129-136, March 1982 J. Max, “Quantizing for minimum distortion”, IEEE Trans. Information Theory, vol.IT-7, pp.7-12, March 1960

  However, the gradation reduction by the linear gradation conversion such as the conventional bit error rounding process has a large image deterioration because the gradation reduction is performed by a simple gradation thinning process at any gradation value. In addition, since the gradation reduction device described in Patent Document 1 weights any image based on the same visual characteristic data, it has not been possible to optimize each image individually.

  On the other hand, the conventional Lloyd-Max method assigns a large number of gradations to gradation values with a high frequency and performs gradation reduction, which is optimal from the viewpoint of minimizing the error between the original image and the gradation-reduced image. This can be said to be a gradation reduction method. However, the Lloyd-Max method cannot be said to be the optimum gradation reduction from the viewpoint of visual and signal processing. This is because, in the Lloyd-Max method, for example, even in a gradation value where noise is dominant, if the frequency is high, gradation is reduced by assigning many gradations.

  An object of the present invention made in view of such circumstances is to provide a gradation reduction device and a program that are excellent in visual and signal processing and that can perform gradation reduction optimized for each image. .

  In order to solve the above-described problem, a gradation reduction apparatus according to the present invention includes a frequency resolving unit that frequency-decomposes an original image to generate a plurality of frequency band components, and an isolated point of the original image by analyzing the frequency band components. A noise-removed image generation unit that generates a noise-removed image in which isolated points are removed from the original image, and a gradation conversion table for generating a histogram of the noise-removed image as training data and reducing gradation A tone reduction initial information generation unit that generates the tone data, and a tone reduction unit that reduces the tone of the original image by the Lloyd-Max method using the training data and the tone conversion table as initial values. Features.

Furthermore, in the gradation reduction device according to the present invention, the noise-removed image generation unit sets the gradation value of a pixel exceeding a first threshold to 1 for the highest frequency band component among the frequency band components, and the first A binarized image in which the gradation value of a pixel equal to or less than a threshold value of 1 is set to 0 is generated, and a pixel having a gradation value of 1 in the binarized image is within a predetermined determination region centered on the pixel. When the total value of the gradation values is equal to or less than the second threshold value, the pixel of the original image corresponding to the pixel is determined as an isolated point .

The gradation reduction apparatus according to the present invention further includes a gradation image generation unit that analyzes the frequency band component to detect a gradation area of the original image and generates a gradation image including only the gradation area. The key reduction initial information generation unit generates a histogram weighted so that the frequency of the gradation value of the gradation image is higher than the histogram of the noise-removed image as the training data .

  Further, the gradation reduction apparatus according to the present invention detects a gradation region of the original image by analyzing the frequency band component by analyzing the frequency band component by frequency-decomposing the original image to generate a plurality of frequency band components, A gradation image generation unit that generates a gradation image consisting only of a gradation region, and a histogram weighted so as to increase the frequency of the gradation value of the gradation image with respect to the histogram of the original image as training data, A gradation reduction initial information generation unit that generates a gradation conversion table for reducing gradation, and the Lloyd-Max method using the training data and the gradation conversion table as initial values, the gradation of the original image is determined. And a gradation reduction unit for reduction.

  Furthermore, in the gradation reduction device according to the present invention, the gradation image generation unit obtains a spatial frequency spectrum at each phase position of the frequency band component to calculate a spatial low frequency power ratio, and the spatial low frequency power A region in which the ratio exceeds a predetermined spectral power threshold is determined as a gradation region.

  In order to solve the above problems, a program according to the present invention causes a computer to function as the gradation reduction device.

  According to the present invention, it is possible to perform gradation reduction which is excellent in visual and signal processing and optimized for each image.

It is a block diagram which shows the structural example of the gradation reduction apparatus of Example 1 of this invention. It is a figure explaining the process by the frequency decomposition part in the gradation reduction apparatus of Example 1 of this invention. It is a block diagram which shows the structural example of the noise removal image generation part in the gradation reduction apparatus of Example 1 of this invention. It is a figure explaining the isolated point determination process by the isolated point detection part in the gradation reduction apparatus of Example 1 of this invention. It is a figure explaining the isolated point removal process by the isolated point detection part in the gradation reduction apparatus of Example 1 of this invention. It is a block diagram which shows the structural example of the gradation reduction initial information generation part in the gradation reduction apparatus of Example 1 of this invention. It is a block diagram which shows the structural example of the gradation reduction apparatus of Example 2 of this invention. It is a block diagram which shows the structural example of the gradation image generation part in the gradation reduction apparatus of Example 2 of this invention. It is a figure explaining the process by the gradation image generation part in the gradation reduction apparatus of Example 2 of this invention. It is a block diagram which shows the structural example of the gradation reduction initial information generation part in the gradation reduction apparatus of Example 2 of this invention. It is a block diagram which shows the structural example of the gradation reduction apparatus of Example 3 of this invention. It is a block diagram which shows the structural example of the gradation reduction initial information generation part in the gradation reduction apparatus of Example 3 of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  First, the gradation reduction apparatus according to the first embodiment of the present invention will be described. The gradation reduction apparatus according to the first embodiment reduces the gradation of the original image in consideration of the noise of the original image. FIG. 1 is a block diagram illustrating a configuration example of the gradation reduction apparatus according to the first embodiment. As shown in FIG. 1, the gradation reduction device 1 includes a frequency decomposition unit 10, a noise-removed image generation unit 20, a gradation reduction initial information generation unit 30, and a gradation reduction unit 40.

  The frequency decomposition unit 10 performs frequency decomposition (for example, wavelet decomposition) on the original image to generate a plurality of frequency band components, and outputs them to the noise-removed image generation unit 20.

  FIG. 2 is a diagram for explaining the frequency resolution processing in the spatial direction by the frequency resolution unit 10. FIG. 2 shows an exploded view when the original image is subjected to third-order wavelet decomposition in the spatial direction. In this exploded view, the spatial frequency component is a higher frequency component on the right side in the horizontal direction and a higher frequency component on the lower side in the vertical direction. As the rank of the wavelet decomposition is increased, the frequency band is more finely decomposed. When the n-th wavelet decomposition is performed, 3n + 1 frequency band components are generated. Note that frequency decomposition may be performed in the time direction in addition to the spatial direction.

  The noise-removed image generation unit 20 detects the isolated point of the original image by analyzing the frequency band component generated by the frequency decomposition unit 10, and regards the isolated point as noise. Then, a noise-removed image from which noise has been removed from the original image is generated and output to the gradation reduction initial information generation unit 30.

  FIG. 3 is a block diagram illustrating a configuration example of the noise removal image generation unit 20. As illustrated in FIG. 3, the noise removal image generation unit 20 includes an isolated point detection unit 201 and a reconstruction unit 202.

The isolated point detection unit 201 analyzes the frequency band component generated by the frequency resolution unit 10 and detects an isolated point of the original image. Specifically, the isolated point detection unit 201 determines the gradation value of a pixel that exceeds a predetermined threshold Th ₁ for the highest frequency band component HH ¹ among the plurality of frequency band components generated by the frequency decomposition unit 10. A binarized image B is generated in which 1 is set and the gradation value of pixels equal to or less than the threshold Th _{1 is} set to 0. Threshold Th ₁ is, for example, the median or average value of the entire gradation values of nonzero frequency band component HH ^1.

The isolated point detection unit 201 calculates, for a pixel whose gradation value is 1 in the binarized image B, a total value of gradation values in a predetermined determination region centered on the pixel and a predetermined threshold Th ₂ . Compare. If the total value of the gradation values of the determination area of the binarized image B exceeds the threshold Th ₂ is the pixel is determined not to be an isolated point elements, the gradation in the determination area of the binarized image B If the total value of the values is the threshold value Th ₂ or less determines determines the pixel as the isolated point element, a pixel of the original image corresponding to the isolated point elements as isolated point.

FIG. 4 is a diagram for explaining isolated point determination processing by the isolated point detection unit 201. In the example shown in FIG. 4, the determination area is 3 × 3 pixels. When the threshold Th ₂ is 1, the gradation value of the binarized image B is 1 only when the gradation values of the 8 pixels around the pixel where the gradation value of the binarized image B is 1 are 0. A pixel is regarded as an isolated element. Therefore, when the threshold Th ₂ = 1, the pixel P1 in the figure is determined to be an isolated point element, and the pixels P2 and P3 are not determined to be isolated point elements.

Then, the isolated point detection unit 201 reconfigures the frequency band component with the gradation value of each frequency band component set to 0 and the gradation value of the isolated point element set to 0 for the pixel position determined to be an isolated point element. To the unit 202. However, since if the noise is included in the DC component it may not be detected as an isolated point element, the frequency band component LL ³ having a DC component is preferable to not change the tone value.

FIG. 5 is a diagram for explaining the isolated point removal processing by the isolated point detection unit 201. When the third-order wavelet decomposition is performed with decimation (that is, the image size of the frequency band component is reduced), 4 × 4 of frequency band components (HH ¹ , HL ¹ , LH ¹ ) whose decomposition rank (decomposition level) is 1 A region of pixels, a 2 × 2 pixel region of frequency band components (HH ² , HL ² , LH ² ) having a decomposition rank of 2, and a frequency band component (HH ³ , HL ³ , LH ³ , ³ ) of a decomposition rank of 3 LL ³ ) corresponds to an area of one pixel. In the drawing, regions A to J at the same phase position are shown. When the isolated point detection unit 201 determines that the pixel j having the frequency band component HH ¹ is an isolated point, the isolated point detection unit 201 sets the gradation value of the region J of 4 × 4 pixels including the pixel j to 0 and the same phase as the region J. The gradation values of the regions B to I at the position are set to 0. It is preferable to not change the gradation value for the area A of the frequency band components LL ³ as described above.

  In order to enable the isolated point detection unit 201 to remove isolated points (noise) in units of pixels of the original image, the frequency resolution unit 10 does not perform decimation (that is, the image size of the frequency band component). Frequency decomposition may be performed with no reduction), and the generated image size of each frequency band component may be the same.

  The reconstruction unit 202 performs reconstruction processing using the frequency band components after detection of isolated points input from the isolated point detection unit 201, generates a noise-removed image, and outputs the noise reduction image to the gradation reduction initial information generation unit 30. . For example, when the frequency decomposition unit 10 performs n-th order wavelet decomposition to generate frequency band components, the frequency band components after detection of isolated points are reconfigured in the n-th order wavelet.

  The gradation reduction initial information generation unit 30 generates a histogram of the noise-removed image generated by the noise-removed image generation unit 20 as training data, and generates a gradation conversion table based on the number of reduced bits. Then, the training data and the gradation conversion table are output to the gradation reduction unit 40. It is assumed that the number of bits to be reduced is specified by the user.

  FIG. 6 is a block diagram illustrating a configuration example of the gradation reduction initial information generation unit 30. As illustrated in FIG. 6, the gradation reduction initial information generation unit 30 includes a histogram generation unit 301 and a gradation conversion table generation unit 302.

  The histogram generation unit 301 generates, as training data, a histogram indicating the frequency for each gradation value for the noise-removed image generated by the noise-removed image generation unit 20, and outputs the training data to the gradation reduction unit 40.

The gradation conversion table generation unit 302 generates a gradation conversion table according to the number of gradation reduction bits and outputs it to the gradation reduction unit 40. The gradation conversion table is a gradation conversion table that linearly converts n bits into (nm) bits when the color depth of the original image is n bits (number of gradations 2 ⁿ ) and the number of gradation reduction bits is m bits. It is.

  The gradation reduction unit 40 generates a gradation-reduced image in which the gradation of the original image is reduced by the Lloyd-Max method using the training data generated by the gradation reduction initial information generation unit 30 and the gradation conversion table as initial values. Generate. The gradation reduction unit 40 determines a quantization step for each gradation according to the frequency of the training data and generates a gradation reduced image, and then performs inverse gradation conversion on the gradation reduced image to obtain the same gradation as the original image. Generate a number of images. Then, the training data and the gradation conversion table are updated and the gradation reduction process is repeated until the difference value between the original image and the image obtained by inverse gradation conversion is equal to or less than a predetermined threshold. According to the Lloyd-Max method, since the quantization step is reduced as the frequency becomes higher, the error between the original image and the gradation-reduced image can be reduced. See Non-Patent Documents 1 and 2 for details of the Lloyd-Max method.

  As described above, the gradation reduction apparatus 1 according to the first embodiment generates a plurality of frequency band components by frequency-decomposing the original image by the frequency decomposition unit 10 and analyzes the frequency band components by the noise removal image generation unit 20. The isolated point of the original image is detected, a noise-removed image from which the isolated point has been removed from the original image is generated, and a histogram of the noise-removed image is generated as training data by the gradation reduction initial information generation unit 30, and gradation is A gradation conversion table for reduction is generated, and the gradation reduction unit 40 reduces the gradation of the original image by the Lloyd-Max method using the training data and the gradation conversion table as initial values.

  As described above, according to the gradation reduction apparatus 1 of the first embodiment, gradation conversion is performed in consideration of noise, and thus gradation reduction that is excellent in visual and signal processing and optimized for each image is performed. be able to. In particular, by smoothing noise when performing gradation reduction, it is possible to reduce subjective image quality degradation due to noise components.

  Note that a computer can be suitably used to cause the above-described gradation reduction apparatus 1 to function. Such a computer stores a program describing the processing contents for realizing each function of the gradation reduction apparatus 1 in a storage unit of the computer, and reads and executes the program by the CPU of the computer. Can be realized.

  Next, the gradation reduction apparatus according to the second embodiment of the present invention will be described. The gradation reduction apparatus according to the second embodiment reduces the gradation of the original image in consideration of the gradation of the original image.

  FIG. 7 is a block diagram illustrating a configuration example of the gradation reduction apparatus according to the second embodiment. As shown in FIG. 7, the gradation reduction device 2 includes a frequency decomposition unit 10, a gradation image generation unit 50, a gradation reduction initial information generation unit 31, and a gradation reduction unit 40. Compared with the gradation reduction device 1 of the first embodiment, the gradation reduction device 2 of the second embodiment includes a gradation image generation unit 50 instead of the noise-removed image generation unit 20, and includes a gradation reduction initial information generation unit 30. The difference is that a gradation reduction initial information generation unit 31 is provided instead.

  Similarly to the first embodiment, the frequency decomposition unit 10 performs frequency decomposition (for example, wavelet decomposition) on the original image to generate a plurality of frequency band components, and outputs the frequency band components to the gradation image generation unit 50.

  The gradation image generation unit 50 detects the gradation region of the original image by analyzing the frequency band component generated by the frequency decomposition unit 10, and generates a gradation image including only the gradation region. Then, the gradation image is output to the gradation reduction initial information generation unit 31.

  FIG. 8 is a block diagram illustrating a configuration example of the gradation image generation unit 50. As shown in FIG. 8, the gradation image generation unit 50 includes a gradation region detection unit 501 and a reconstruction unit 502.

  The gradation area detection unit 501 acquires the spatial frequency spectrum of each phase position of the frequency band component generated by the frequency decomposition unit 10 and obtains the ratio of the spatial low frequency power at the same phase position (the spatial low frequency relative to the spectrum of the entire frequency band). The ratio of the spectrum in the frequency band) is calculated, and an area where the ratio of the spatial low frequency power exceeds a predetermined spectrum power threshold is determined as the gradation area. Then, an image including only the detected gradation area is output to the reconstruction unit 502.

  As the number of gradation reduction bits is larger, the area where the gradation changes naturally becomes an area where the gradation changes stepwise, and the possibility that a pseudo contour is generated increases. And in the area | region where a gradation changes stepwise, the ratio of spatial high frequency power increases. For this reason, it is preferable that the gradation area detection unit 501 determines that the spectrum power threshold is smaller as the number of gradation reduction bits is larger, thereby increasing the detection accuracy of the gradation area.

  Further, the gradation area detection unit 501 may store a correspondence table in which the number of gradation reduction bits and the spectrum power threshold are associated in advance, and determine the spectrum power threshold based on the correspondence table. Table 1 shows an example of a correspondence table between the number of gradation reduction bits and the spectrum power threshold. The spectrum power threshold is 1 at maximum.

FIG. 9 is a diagram for explaining processing by the gradation area detection unit 501. When the third-order wavelet decomposition is performed with decimation (that is, the image size of the frequency band component is reduced), 4 × 4 of frequency band components (HH ¹ , HL ¹ , LH ¹ ) whose decomposition rank (decomposition level) is 1 A region of pixels, a 2 × 2 pixel region of frequency band components (HH ² , HL ² , LH ² ) having a decomposition rank of 2, and a frequency band component (HH ³ , HL ³ , LH ³ , ³ ) of a decomposition rank of 3 LL ³ ) corresponds to an area of one pixel. In the drawing, regions A to J at the same phase position are shown. When the ratio of the power of the area A of the frequency band component LL ³ is obtained, the ratio of the power of the area A to the areas A to J is obtained. When the power ratio of the region A is equal to or less than the spectral power threshold, the gradation values of the regions B to I in the same phase position as the region A are set to 0.

  In order to enable the gradation area detection unit 501 to detect the gradation area in units of pixels of the original image, the frequency resolution unit 10 does not perform decimation (that is, there is no reduction in the image size of the frequency band component). The frequency resolution may be performed in step S1 to generate the same image size for each frequency band component generated.

  The reconstruction unit 502 performs reconstruction processing using the frequency band component after gradation region detection input from the gradation region detection unit 501, generates a gradation image, and outputs the gradation image to the gradation reduction initial information generation unit 30. For example, when the frequency decomposition unit 10 performs n-th order wavelet decomposition to generate frequency band components, the frequency band components after gradation region detection are reconfigured to the nth order wavelet.

  The gradation reduction initial information generation unit 31 generates training data based on the original image and the gradation image generated by the gradation image generation unit 50, and also generates a gradation conversion table based on the number of reduction bits. Then, the training data and the gradation conversion table are output to the gradation reduction unit 40.

  FIG. 10 is a block diagram illustrating a configuration example of the gradation reduction initial information generation unit 31. As illustrated in FIG. 10, the gradation reduction initial information generation unit 31 includes a histogram generation unit 311, a weighting unit 312, and a gradation conversion table generation unit 313.

  The histogram generation unit 311 generates a histogram indicating the frequency for each gradation value for the original image, and outputs the histogram to the weighting unit 312.

  The weighting unit 312 detects the gradation value of the gradation image generated by the gradation image generation unit 50 and performs weighting so that the frequency of the gradation image is higher than the histogram of the original image generated by the histogram generation unit 311. To do. Then, the weighted histogram is output to the gradation reduction unit 40 as training data. For example, the frequency of the histogram of the gradation image is changed to a value multiplied by a value exceeding a predetermined value 1 (for example, 1.2). It should be noted that the higher the frequency, the higher the weight. By using the weighted histogram as training data, more gradations can be assigned to the gradation area when the gradation reduction unit 40 performs gradation reduction.

The gradation conversion table generation unit 313 generates a gradation conversion table according to the number of gradation reduction bits and outputs it to the gradation reduction unit 40. The gradation conversion table is a gradation conversion table that linearly converts n bits into (nm) bits when the color depth of the original image is n bits (number of gradations 2 ⁿ ) and the number of gradation reduction bits is m bits. It is.

  Similar to the first embodiment, the gradation reduction unit 40 uses the training data generated by the gradation reduction initial information generation unit 31 and the gradation conversion table as an initial value to perform the gradation of the original image by the Lloyd-Max method. A gradation-reduced image with reduced image is generated.

  As described above, the gradation reduction apparatus 2 according to the second embodiment generates a plurality of frequency band components by frequency-decomposing the original image by the frequency decomposition unit 10 and analyzes the frequency band components by the gradation image generation unit 50. The gradation area of the original image is detected, a gradation image consisting only of the gradation area is generated, and the gradation reduction initial information generation unit 31 causes the gradation value of the gradation image to be higher than the histogram of the original image. Is generated as training data, a gradation conversion table for reducing gradation is generated, and the gradation reduction unit 40 uses the training data and the gradation conversion table as initial values by the Lloyd-Max method. Reduce the gradation of the original image.

  As described above, according to the gradation reduction device 2 of the second embodiment, gradation conversion is performed in consideration of gradation, so that gradation reduction that is excellent in visual and signal processing and optimized for each image is performed. be able to. In particular, by assigning a large number of gradations to gradation gradations when performing gradation reduction, it is possible to suppress the occurrence of pseudo contours during gradation restoration.

  Note that a computer can be suitably used in order to function as the gradation reduction device 2 described above. Such a computer stores a program describing processing contents for realizing each function of the gradation reduction device 2 in a storage unit of the computer, and reads and executes the program by the CPU of the computer. Can be realized.

  Next, a gradation reducing apparatus according to Embodiment 3 of the present invention will be described. The gradation reduction apparatus according to the third embodiment reduces the gradation of the original image in consideration of noise and gradation of the original image.

  FIG. 11 is a block diagram illustrating a configuration example of the gradation reducing apparatus according to the third embodiment. As shown in FIG. 11, the gradation reduction device 3 includes a frequency decomposition unit 10, a noise-removed image generation unit 20, a gradation image generation unit 50, a gradation reduction initial information generation unit 32, and a gradation reduction unit 40. And comprising. The gradation reduction device 3 according to the third embodiment further includes a gradation image generation unit 50 as compared with the gradation reduction device 1 according to the first embodiment, and replaces the gradation reduction initial information generation unit 30 with gradation reduction initial information. The point provided with the production | generation part 32 is different.

  Similarly to the first embodiment, the frequency decomposition unit 10 performs frequency decomposition (for example, wavelet decomposition) on the original image to generate a plurality of frequency band components. Then, the frequency band component is output to the noise removal image generation unit 20 and the gradation image generation unit 50.

  Similar to the first embodiment, the noise-removed image generation unit 20 analyzes the frequency band component generated by the frequency decomposition unit 10 to detect an isolated point of the original image, and regards the isolated point as noise. Then, a noise-removed image from which noise has been removed from the original image is generated and output to the gradation reduction initial information generation unit 32.

  As in the second embodiment, the gradation image generation unit 50 analyzes the frequency band component generated by the frequency decomposition unit 10 to detect the gradation area of the original image, and generates a gradation image including only the gradation area. Then, the gradation image is output to the gradation reduction initial information generation unit 32.

  The gradation reduction initial information generation unit 32 generates training data based on the noise-removed image generated by the noise-removed image generation unit 20 and the gradation image generated by the gradation image generation unit 50, and reduces the number of bits to be reduced. A gradation conversion table is generated based on the gradation conversion table. Then, the training data and the gradation conversion table are output to the gradation reduction unit 40.

  FIG. 12 is a block diagram illustrating a configuration example of the gradation reduction initial information generation unit 32. As illustrated in FIG. 12, the gradation reduction initial information generation unit 32 includes a histogram generation unit 321, a weighting unit 322, and a gradation conversion table generation unit 323.

  The histogram generation unit 321 generates a histogram indicating the frequency for each gradation value for the noise-removed image generated by the noise-removed image generation unit 20, and outputs the histogram to the weighting unit 322.

  The weighting unit 322 detects the gradation value of the gradation image generated by the gradation image generation unit 50, and the frequency of the gradation value of the gradation image is determined with respect to the histogram of the noise-removed image generated by the histogram generation unit 321. Weight to be higher. Then, the weighted histogram is output to the gradation reduction unit 40 as training data. For example, the frequency of the gradation value of the gradation image is changed to a value obtained by multiplying a predetermined value exceeding 1 (for example, 1.2). Note that the weight may be increased as the gradation value of the gradation image having a higher frequency. By using the weighted histogram as training data, more gradations can be assigned to the gradation area when the gradation reduction unit 40 performs gradation reduction.

The gradation conversion table generation unit 323 generates a gradation conversion table according to the number of gradation reduction bits and outputs it to the gradation reduction unit 40. The gradation conversion table is a gradation conversion table that linearly converts n bits into (nm) bits when the color depth of the original image is n bits (number of gradations 2 ⁿ ) and the number of gradation reduction bits is m bits. It is.

  Similar to the first embodiment, the gradation reduction unit 40 uses the training data generated by the gradation reduction initial information generation unit 32 and the gradation conversion table as an initial value to perform the gradation of the original image by the Lloyd-Max method. A gradation-reduced image with reduced image is generated.

  As described above, the gradation reduction device 3 according to the third embodiment generates a plurality of frequency band components by frequency-decomposing the original image by the frequency decomposition unit 10, and analyzes the frequency band components by the noise removal image generation unit 20. An isolated point of the original image is detected, a noise-removed image is generated by removing the isolated point from the original image, a gradation region of the original image is detected by analyzing a frequency band component by the gradation image generation unit 50, and the gradation region A gradation image consisting only of the above, and the gradation reduction initial information generation unit 32 generates a histogram weighted so that the frequency of the gradation value of the gradation image is higher than the histogram of the noise-removed image as training data, A gradation conversion table for reducing gradation is generated, and training data and gradation conversion table are generated by the gradation reduction unit 40. The Lloyd-Max method using as an initial value, reducing the gray scale of the original image.

  As described above, according to the gradation reduction device 3 of the third embodiment, the gradation conversion is performed in consideration of the noise region and gradation, so that the gradation is excellent in visual and signal processing and optimized for each image. Reductions can be made. In particular, when performing gradation reduction, smoothing noise and assigning many gradations to gradation gradations reduces subjective image quality degradation due to noise components, and also generates pseudo contours during gradation restoration. Can be suppressed.

  Note that a computer can be suitably used to function as the gradation reduction device 3 described above, and such a computer can store a program that describes the processing contents for realizing each function of the gradation reduction device 3. This program can be realized by reading out and executing the program by the CPU of the computer.

  Each of the above embodiments has been described as a representative example, but it will be apparent to those skilled in the art that many changes and substitutions can be made within the spirit and scope of the invention. Therefore, the present invention should not be construed as being limited by the above-described embodiments, and various modifications and changes can be made without departing from the scope of the claims. For example, in the embodiment, wavelet decomposition is used as frequency decomposition, but the present invention is not limited to this.

  Thus, the present invention is useful for any application that reduces gradation. For example, to reduce the gradation of an image that has a large noise component when a sufficient amount of light cannot be obtained due to high resolution, such as Super Hi-Vision, or to greatly reduce the gradation from 12-bit gradation to 8-bit gradation or less. It is particularly useful for applications.

1, 2, 3 Gradation reduction device 10 Frequency decomposition unit 20 Noise removal image generation unit 30, 31, 32 Gradation reduction initial information generation unit 40 Gradation reduction unit 50 Gradation image generation unit 201 Isolated point detection unit 202 Reconstruction unit 301, 311, 321 Histogram generation unit 302, 313, 323 Gradation conversion table generation unit 312, 322 Weighting unit 501 Gradation area detection unit 502 Reconfiguration unit

Claims (6)

A frequency resolving unit that generates a plurality of frequency band components by frequency resolving the original image;
A noise-removed image generation unit that analyzes the frequency band component to detect an isolated point of the original image and generates a noise-removed image in which the isolated point is removed from the original image;
A tone reduction initial information generating unit that generates a histogram of the noise-removed image as training data and generates a tone conversion table for reducing tone,
A gradation reduction unit that reduces the gradation of the original image by the Lloyd-Max method using the training data and the gradation conversion table as initial values;
A gradation reduction device comprising: The noise-removed image generation unit sets a gradation value of a pixel exceeding a first threshold value to 1 and a gradation value of a pixel equal to or less than the first threshold value to 0 for the highest frequency band component among the frequency band components. For a pixel having a gradation value of 1 in the binarized image, the total value of the gradation values in a predetermined determination area centered on the pixel is equal to or less than a second threshold value. If it is is characterized by determining a pixel of the original image corresponding to the pixel an isolated point, the gradation reduction apparatus according to claim 1. Further comprising: a gradation image generating unit that analyzes the frequency band component to detect a gradation area of the original image and generates a gradation image including only the gradation area;
The gradation reduction initial information generation unit generates, as the training data, a weighted weight so that the gradation value of the gradation image becomes higher than the histogram of the noise-removed image . The gradation reduction apparatus according to claim 1 or 2 . A frequency resolving unit that generates a plurality of frequency band components by frequency resolving the original image;
A gradation image generating unit that analyzes the frequency band component to detect a gradation region of the original image and generates a gradation image including only the gradation region;
A gradation that generates a gradation conversion table for reducing gradation while generating a histogram weighted so that the frequency of gradation values of the gradation image is higher than the histogram of the original image. A reduction initial information generation unit;
The Lloyd-Max method using the training data and the gradation conversion table as initial values, a gradation reducing unit that reduces the gradation of the original image,
A gradation reduction device comprising: The gradation image generation unit obtains a spatial frequency spectrum of each phase position of the frequency band component to calculate a spatial low frequency power ratio, and an area where the spatial low frequency power ratio exceeds a predetermined spectral power threshold. The gradation reduction device according to claim 3 or 4 , wherein the gradation reduction device is determined as a gradation region.   The program for functioning a computer as a gradation reduction apparatus as described in any one of Claim 1 to 5.

Images